select search filters
roundups & rapid reactions
factsheets & briefing notes
before the headlines
Fiona fox's blog

coffee consumption and reduced risk of death

Two studies publishing in the Annals of Internal Medicine assess the association between coffee consumption and risk of death.


A roundup accompanied this analysis.


Title, Date of Publication & Journal

Coffee drinking and mortality in 10 European countries: A multinational cohort study by Mark J. Gunter et al.

Published: July 10th 2017

Annals of Internal Medicine 


Study’s main claims – and are they supported by the data

The main claim is that coffee drinking was associated with a reduced risk of death from any cause.  This is partly, but not entirely, supported by the evidence – see below.

This type of study is not able to demonstrate that coffee is responsible for (i.e. causes) a reduced risk of death, either for overall death or specific causes of death (e.g. cancer, heart disease).  However, the Imperial College London press release headline – ‘Drinking coffee reduces risk of death from all causes, study finds’, makes this claim.  This is not what the study found, nor what the authors imply.

The study also investigated whether coffee was associated with specific death types, such as cancer, heart disease, or digestive diseases.  The study found a positive association between higher coffee consumption and risk of cancer death amongst women (higher coffee consumption associated with higher cancer mortality in women).  The press release focuses only on negative associations (reduced mortality with higher coffee consumption).  The study discussion downplays this finding, suggesting it ‘may be spurious’, but they do not apply similar scepticism to the study’s other findings.

Further, the authors say “higher levels of coffee drinking are associated with lower risk for death from various causes”, but this ignores that risk of death appears to be greater in those who drink a high amount of coffee compared with those who drink a medium amount of coffee.




It is a large study population, which appears to be fairly representative of healthy western Europeans in the 1990s over 35 years old (see first limitation below).

Appropriately stratified by country, age and sex.


Participants who reported cancer, heart disease, stroke, or diabetes at recruitment were excluded.  The authors do not explain or justify this decision.  Did these participants have similar coffee consumption habits to those participants who were included in the study?  If not, then there is a substantial selection bias introduced to the study which could have exaggerated the already quite modest findings.  This makes the results difficult to interpret as, for instance, a large group of coffee drinkers whose present ill-health was related to coffee may have been excluded.

Coffee consumption was assessed only once, at the time participants were recruited to the study.  This was not updated throughout the study period and so the study does not account for changes in coffee consumption habits.  Therefore, any reference to levels of coffee consumption are based on one point in time that may not accurately reflect the participants’ future (or former) coffee habits.

Although the authors used statistical methods to give some idea of the relative difference in risk of death in high and low coffee drinkers, they don’t give absolute numbers so it’s difficult to gauge the size of the effect.

The conclusions ignore the non-linear trends suggested in Table 2, with hazard ratios for Quartile 4 consistently higher than hazard ratios for Quartile 3 – in other words, participants who consume a high amount of coffee are associated with a higher risk of death compared with those who consume a medium amount of coffee.

The follow-up period is on average about half a generation (mean follow up of 16.4 years).  This is small compared to other large, national cohort studies.  The results may be driven more by older participants who were more likely to die during the study (and who were found to drink less coffee than younger participants on average).

As there were lots of different analyses for different causes of death and for men and women separately, the risk of finding false positive results is increased.


Any specific expertise relevant to studied paper (beyond statistical)?



* European study: ‘Coffee drinking and mortality in 10 European countries: A multinational cohort study’ by Marc J. Gunter et al. published in the Annals of Internal Medicine at on Monday 10 July 2017.

American study: ‘Association of coffee consumption with total and cause-specific mortality among nonwhite populations’ by Song-Yi Park et al. published in the Annals of Internal Medicine on Monday 10 July 2017.

Editorial: ‘Moderate coffee intake can be part of a healthy diet’ by Eliseo Guallar et al. published in the Annals of Internal Medicine on Monday 10 July 2017.


Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.

in this section

filter Headlines by year

search by tag